Data Visualisation

Data Visualisation

Du skal tegne før du må regne / “You must illustrate before you calculate”
— George Rasch or Thorvald Nicolai Thiele

A picture is worth a thousand words.
— Henrik Ibsen

• Data visualisation communicates information much quicker than numerical tables
• Data visualisation can reveal unexpected structures in data (a key tools in exploratory data analysis)

Making the correct visual summary of a dataset or the results of an analysis is the most powerful and conveying method to present your work to others.

Hidden structures

# A tibble: 142 × 2
       x     y
   <dbl> <dbl>
 1  55.4  97.2
 2  51.5  96.0
 3  46.2  94.5
 4  42.8  91.4
 5  40.8  88.3
 6  38.7  84.9
 7  35.6  79.9
 8  33.1  77.6
 9  29.0  74.5
10  26.2  71.4
# ℹ 132 more rows

Hidden structures

Data from Alberto Cairo.

The Datasaurus Dozen

Image credit: Steph Locke, based on datasauRus / DOI:10.1145/3025453.3025912.

Programming graphics

An iterative process

The ggplot2 package

• R has several ways to plot and visualise data and models. We typically group these in two overall categories - base R and ggplot.

• The former is the built-in feature in R that historically made R a favorite tool for visualisation compared to other contemporary statistic software (e.g. Excel, SAS, SPSS, etc.).

• However, with the introduction of the ggplot2 package in 2007 the features and functionalities in R’s visual toolbox has increased dramatically.

• ggplot2 implements what is referred to as “The Grammar of Graphics”. Here each feature in a plot is controlled by a variable in the data.

A graphing template for layered plots

ggplot(data = <DATA>, mapping = aes(<MAPPINGS>)) +
  layer(geom = <GEOM>, stat = <STAT>, position = <POSITION>)

Fuel economy data (miles per gallon) from ggplot2:

mpg

# A tibble: 234 × 11
   manufacturer model  displ  year   cyl
   <chr>        <chr>  <dbl> <int> <int>
 1 audi         a4       1.8  1999     4
 2 audi         a4       1.8  1999     4
 3 audi         a4       2    2008     4
 4 audi         a4       2    2008     4
 5 audi         a4       2.8  1999     6
 6 audi         a4       2.8  1999     6
 7 audi         a4       3.1  2008     6
 8 audi         a4 qu…   1.8  1999     4
 9 audi         a4 qu…   1.8  1999     4
10 audi         a4 qu…   2    2008     4
# ℹ 224 more rows
# ℹ 6 more variables: trans <chr>,
#   drv <chr>, cty <int>, hwy <int>,
#   fl <chr>, class <chr>

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  layer(geom = "point", stat = "identity", position = "identity")

layer() short-hands:

• geom_point() shorthand for
• layer(geom = "point", stat = "identity", position = "identity")

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  layer(geom = "point", stat = "identity", position = "identity")

ggplot(mpg, aes(displ, hwy)) + 
  geom_point()

• Often we will use the pipe to start with data and send it to ggplot-function:

mpg |> ggplot(aes(displ, hwy)) + 
  geom_point()

Examples

mpg |> ggplot(aes(x = displ, y = hwy))
mpg |> ggplot(aes(x = displ, y = hwy)) + 
  geom_point()
mpg |> ggplot(aes(x = displ, y = hwy)) + 
  geom_point(aes(colour = class))

Inheriting mappings and data

• We have a structure where the data is defined and features of the plot is controlled by adding layers to the plot

ggplot(data = <DATA>) +
  <GEOM_FUNCTION>(mapping = aes(<MAPPINGS>))

ggplot(data = <DATA>, mapping = aes(<MAPPINGS>)) +
  <GEOM_FUNCTION>()

• Note, if an aesthetic is set by the mapping at the ggplot-level it will be inherited by all the subsequent layers (globally). Similarly for data.

mpg |> ggplot(aes(x = displ, y = hwy, colour = class)) + 
  geom_point() + 
  geom_smooth()

• However, if specified at a GEOM_FUNCTION it will only affect that geom (locally).

mpg |> ggplot(aes(x = displ, y = hwy)) + 
  geom_point(aes(colour = class)) + 
  geom_smooth()

Controllable aesthetics

Which features/aesthetics that can be controlled depends on the type of plot, e.g.:

• scatterplot
  • colour
  • size
  • shape
  • alpha
  • fill (only for specific plot characters)
  • stroke (only for specific plot characters)
• boxplot
  • fill (of the boxes)
  • color (outline colour)
  • …
• barplot
  • fill (of the boxes)
  • color (outline colour)
  • …
• histogram
  • fill (of the boxes)
  • color (outline colour)
  • …
• maps
• heatmaps
• densities
• …

The help packages for each geom, e.g. ?geom_point tell which elements that can be controlled.

Facets

Another powerful feature of ggplot2 is the ease of making facet plots (subplots where each panel/plot is conditioned on other variables)

mpg_plot <- mpg |> ggplot(aes(x = displ, y = hwy)) + 
  geom_point(aes(colour = class)) + 
  geom_smooth(method = "lm", formula = y ~ x)

mpg_plot + facet_wrap(~manufacturer)

mpg_plot + facet_wrap(~manufacturer, scales = "free_x")

mpg_plot + facet_wrap(~manufacturer, scales = "free")

One can facet on several variables at the same time

## Construct a new variable from `trans` to determine Auto or Manual transmisson:
## NB! When making changes to the data we need to "revoke" ggplot once again for it to work!
mpg |> 
  mutate(transmission = ifelse(str_detect(trans, "auto"), "Auto", "Manual")) |> 
  ggplot(aes(x = displ, y = hwy)) + 
  geom_point(aes(colour = class)) + 
  geom_smooth(method = "lm", formula = y ~ x) + 
  facet_grid(transmission ~ cyl, labeller = label_both)

Position adjustment

When plotting histograms, barplots and boxplots it is important to be able to control the position of the graphical elements.

There is a “family” of position-functions that can be used to make finer control and adjustment of the position:

• position_dodge
• position_dodge2
• position_fill
• position_identity
• position_jitter
• position_jitterdodge
• position_nudge
• position_stack

The online help pages has many examples: https://ggplot2.tidyverse.org/reference/position_dodge.html

Example: Barplot

mpg2 <- mpg |> mutate(transmission = ifelse(str_detect(trans, "auto"), "Auto", "Manual"))

• What is wrong with the fill color here?

mpg2 |> 
  ggplot(aes(x = transmission, fill = cyl)) + 
  geom_bar()

mpg2

# A tibble: 234 × 12
   manufacturer model    displ  year   cyl
   <chr>        <chr>    <dbl> <int> <int>
 1 audi         a4         1.8  1999     4
 2 audi         a4         1.8  1999     4
 3 audi         a4         2    2008     4
 4 audi         a4         2    2008     4
 5 audi         a4         2.8  1999     6
 6 audi         a4         2.8  1999     6
 7 audi         a4         3.1  2008     6
 8 audi         a4 quat…   1.8  1999     4
 9 audi         a4 quat…   1.8  1999     4
10 audi         a4 quat…   2    2008     4
# ℹ 224 more rows
# ℹ 7 more variables: trans <chr>,
#   drv <chr>, cty <int>, hwy <int>,
#   fl <chr>, class <chr>,
#   transmission <chr>

Example: Barplot

cyl was numeric

mpg2 |> ggplot(aes(x = transmission, fill = factor(cyl))) + geom_bar()

mpg2 |> ggplot(aes(x = transmission, fill = factor(cyl))) + 
  geom_bar() + labs(fill = "Cylinders")

Example: Barplot

Not stacked - side-by-side

mpg2 |> ggplot(aes(x = transmission, fill = factor(cyl))) + 
  geom_bar(position = position_dodge()) + labs(fill = "Cylinders")

Example: Barplot

Relative distribution within each group

mpg2 |> ggplot(aes(x = transmission, fill = factor(cyl))) + 
  geom_bar(position = position_fill()) + 
  labs(y = "fraction", fill = "Cylinders")

Difference between geom_bar and geom_col

Barcharts/plots are often used to present summarised data by different combinations of qualitative variables.

In ggplot2 there are two geoms that do almost the same: geom_bar and geom_col. The former performs a statistical calculation on the data before plotting - the latter doesn’t.

Hence, if we already have aggregated data we use geom_col and otherwise geom_bar.

Themes

In ggplot2 there are a number of themes that controls the overall appearance of the plot. There are a number of themes that comes with ggplot2

• theme_bw
• theme_classic
• theme_dark
• theme_gray /theme_grey
• theme_light
• theme_linedraw
• theme_minimal
• theme_void

The ggthemes provides even more:

• theme_base
• theme_calc
• theme_clean
• theme_economist
• theme_economist_white
• theme_excel
• theme_excel_new
• theme_few
• theme_fivethirtyeight
• theme_foundation
• theme_gdocs
• theme_hc
• theme_igray
• theme_map
• theme_pander
• theme_par
• theme_solarized
• theme_solarized_2
• theme_solid
• theme_stata
• theme_tufte
• theme_wsj

The function theme_set can be used to set the theme globally in a session:

theme_set(theme_bw(base_size = 20))

Use theme for finer control

The theme-function enables an even finer control of the look.

We typically use the theme to control the position of e.g. the legend.

mpg2 |> ggplot(aes(x = displ, y = hwy, colour = transmission)) + 
  geom_point() + geom_smooth(method = "lm", formula = y ~ x) + 
  theme(legend.position = "top")

Typical error

NB! Remember to use theme before adding a change to the theme

mpg2 |> ggplot(aes(x = displ, y = hwy, colour = transmission)) + 
  geom_point() + geom_smooth(method = "lm", formula = y ~ x) + 
  theme(legend.position = "top") + theme_excel()

mpg2 |> ggplot(aes(x = displ, y = hwy, colour = transmission)) + 
  geom_point() + geom_smooth(method = "lm", formula = y ~ x) + 
  theme_excel() + theme(legend.position = "top") 

Vast list of resources

Stack overflow: https://stackoverflow.com/questions/tagged/ggplot2 (Over 50k questions)

Types of geoms: https://ggplot2.tidyverse.org/reference/index.html

Gallery: https://exts.ggplot2.tidyverse.org/gallery/

Cheatsheet: https://github.com/rstudio/cheatsheets/blob/main/data-visualization-2.1.pdf

Exercises

Try to have a go with some of the many exercises in Chapter 1 (also available in qmd file on Moodle):

• Exercises 1.2.5: https://r4ds.hadley.nz/data-visualize#exercises
• Exercises 1.4.3: https://r4ds.hadley.nz/data-visualize#exercises-1
• Exercises 1.5.5: https://r4ds.hadley.nz/data-visualize#exercises-2